Land Management and Technology Diffusion in Uganda
PhD progress report
By: Johannes Woelcke
December 2001
Zentrum für Entwicklungsforschung
Center for Development Research
Universität Bonn
Executive summary
Land degradation is a major concern in Uganda contributing to declining agricultural
productivity, poverty and food insecurity. In addition, a fast growing population raises the
need for growth of agricultural output. Increased crop output will have to come from higher
yields as the arable land frontier is closing. Therefore, an important research issue is how to
assist policy makers in designing policy interventions that contribute to a sustainable
intensification of agriculture. Technologies, which have the potential to reach the objectives
of increasing productivity and sustainability simultaneously, have to be identified and
promoted. An appropriate instrument to address these problems is a bioeconomic model,
which combines socio-economic factors influencing farmers` objectives and constraints with
biophysical factors affecting production possibilities and the impact of land management
practices.
1. Introduction
1.1 General background
A widening gap between food production and food needs is observed in many developing
countries. Urgently needed increased crop output will have to come from higher yields largely
because the arable land frontier is closing in most developing countries. Intensification of
agriculture will transform the environment. Hence, an important issue is whether, and how,
agricultural growth can be attained simultaneously with the conservation of the natural
resources of farmland.
In the developing countries one of the most serious environmental threats is land degradation
contributing to declining agricultural productivity, poverty and food insecurity. In the journal
“Land degradation and development” (2000) land degradation is defined “as the loss of
utility through reduction of or damage to physical, social or economic features and/or
reduction of ecosystem diversity.”
The “critical triangle of development goals” (Vosti and Reardon, 1997) implies that it is a
major objective for researchers and politicians to find policies, institutions and technologies to
make the three goals growth, poverty alleviation and sustainability more compatible. It is
obvious that the three goals are compatible in the long run. Sustaining the natural resource
base will help agricultural productivity growth and this will lead to poverty alleviation. In the
short run there might be trade-offs among the three goals taking into account the short-term
perspective of the individual farmer to satisfy the basic needs of the household.
Farmers need to have the incentive and the capacity for a sustainable intensification of
agriculture. Several factors such as policies, technologies, institutions, population pressure
and agroclimatic conditions can affect the links between sustainability, growth and poverty
alleviation by influencing the choices of households and communities. These factors have the
potential to increase the compatibility of the three objectives.
1.2 Agriculture in Uganda
Under the regimes of Idi Amin (1971-79) and Milton Obote (1980-1985) Uganda’s economy
plunged into a prolonged crisis with negative real GDP growth rates (Baffoe, 2000).
In 1987 the Ugandan Government under Musevini introduced an Economy Recovery Program
in cooperation with the IMF and World Bank aiming at market liberalization, privatisation
and decentralization. Although, these reforms have some positive impacts on Ugandan
economy (GDP real growth has averaged 6 per cent per annum), the productivity in the
agricultural sector has either stagnated or declined (APSEC, 2000). Therefore, the Ugandan
Government has published a “Plan for Modernization of Agriculture” (Government of the
Republic of Uganda, 2000) as part of the “Poverty Eradication Action Plan (PMA)” with the
vision of “poverty eradication through a profitable, competitive, sustainable and dynamic
agricultural and agro-industrial sector.” The priority areas for action are: improving access to
rural finance, improving access to markets, research and technology development, sustainable
natural resource utilization and management and education for agriculture.
The mainstay of Uganda’s economy is the agricultural sector, which accounts for 43 % of the
GDP, 85 % of the value of exports and which provides 80 % of employment (FAO, 1999).
Uganda’s agriculture is dominated by food crop production, which contributes more than two
thirds to agricultural GDP. Livestock accounts for another 23 %. In 1993 eighty-nine percent
of the population were rural, agricultural output came almost exclusively from about 2,5
million smallholders, 80 % of whom have less than 2 hectares each (World Bank, 1993).
In Uganda, as in many other countries in sub-Saharan Africa, land degradation problems are a
growing concern leading to declining agricultural productivity and poverty.
Per capita agricultural production and crop yields per unit area of production is declining in
Uganda (Sanchez et al. 1996). The soils were once considered to be among the most fertile in
the tropics (Chenery, 1960), but recently land degradation appears to be increasing. It was
estimated that soil nutrient losses were among the highest in sub-Saharan Africa in the early
80s and average annual nutrient losses were predicted to reach 85 kg/ha of N, P2O5, K2O by
the year 2000 (Stoorvogel and Smaling 1990).
Recent studies in eastern and central Uganda have given negative nutrient balances for most
of the cropping systems (Wortmann and Kaizzi 1998).
The proximate causes of nutrient depletion are very low use of inorganic fertilizers and
limited use of organic inputs coupled with declining fallow periods. The proximate causes of
soil erosion are deforestation, crop production on steep slopes with limited investments in
terraces or other conservation measures.
1.3 Problem statement
As mentioned above the proximate causes of land degradation are relatively well known but
the core of the land degradation problem is economic. Poor rural households in developing
countries have to cope with a situation where land productivity and therefore household
income are stagnant or declining. Financial constraints and imperfect capital markets are
leading to livelihood strategies that contribute to nutrient depletion since sustainable
intensification of agriculture is not affordable for the majority of farm households (Barbier,
1997). A critical research challenge that has not been solved yet is to improve understanding
of the key factors affecting land management and to assess the impacts of policy interventions
and alternative technology. It is a key challenge to identify technologies that simultaneously
meet growth and sustainability goals.
Another important and difficult task is to design effective policy strategies to make these
technologies affordable and adoptable for the farmers, including poor farmers. A lot of studies
(Feder, Just and Zilberman, 1985) have been conducted analysing the determinants, which
influence the adoption of technology (i.e. farm size, tenure, age, education and risk), but how
farm households react to alternative policy strategies and how the adoption of a technology
affects the environment and the productivity simultaneously is less clear.
2. Research objectives
Consideration of the problem presented above led to the following research objectives:
1. Improve understanding of key economic determinants affecting land management
decisions at the farm household level;
2. Better understanding of the diffusion patterns of promoted agricultural technologies
and their impact on household welfare and the condition of natural resources;
3. Draw conclusions for the design of land use policies, which promote more productive,
sustainable, and poverty-reducing land management in Uganda.
With respect to research objective 1 the following hypothesis will be tested:

Labor shortages, capital constraints, imperfect capital markets, distorted input and
output prices, transaction and information costs are the most binding factors affecting
land use practices and adoption of new technologies.
With respect to research objective 2 the following hypotheses will be tested:

Social networks are a critical factor for the adoption/diffusion of innovations. To
realize widespread technology adoption a critical mass has to be reached through
providing incentives for early adoption (extension services, reduced prices etc.) and
introducing the innovation to opinion leaders (the critical mass occurs at the point at
which enough individuals adopt an innovation so that it’s further rate of adoption
becomes self sustaining, Rogers 1995).

“Overlap technologies” exist which simultaneously meet growth and sustainability
goals, but missing incentives and physical and financial constraints prohibit farmers
from adopting these technologies;
The following scenario will be developed:

Prediction of likely patterns of promoted technologies and their impact on household
welfare and the condition of natural resources
With respect to research objective 3 the following scenarios will be developed:

Effects of market liberalization policies (development of input and output prices,
subsidies and interest rates) on productivity and sustainability criteria

Policy and institutional interventions mentioned as priority areas in the PMA
(development of local credit markets, public investments in infrastructure, labor
exchange institutions etc.), which affect farmers` choices of land management
practices.
3. Conceptual framework
Farm household models offer a promising perspective for the analysis of production and
consumption decisions at the farm level (Singh et al., 1986). Farm households are considered
to be the central decision makers regarding agricultural production. Individual farmers have to
decide which commodities to produce in which quantities, by which method, in which
seasonal time periods. It is the objective of the farmers to maximize their utility, which
deviates from pure profit maximizing behaviour in many cases. For example, risk aversion,
leisure and providing enough food for household consumption are important goals, which
have to be taken into account, too. The decision-making procedure is subject to physical and
financial constraints (e.g., acres of land, days of labor and limited credit availability) as well
as uncertainty about the next planning periods. Uncertainty arises for example, in forecasted
yields, costs and prices. Linkages between production and consumption decisions,
characteristic for farm households operating under imperfect markets, have to be included.
Due to the possibility of analysing both, production and consumptions decisions, the farm
household model approach represents a useful starting point for the analysis of effectiveness
of economic policy instruments to enhance a sustainable intensification of land management.
Policy analysis for sustainable land use proves to be critically dependent on the specification
of the linkages between decision-making procedures regarding resource allocation by farm
households and their supply response to changes in the economic, institutional and ecological
environment. Decisions of the farm household are influenced by external factors. These
decisions have in turn consequences for the agricultural production and the conditions of
natural resource conditions.
The agro-ecological and socio-economic environment are considered to be the most important
external factors determining farm household decision-making. The agro-ecological
environment defines the potential agricultural production activities from which the
households can select. The socio-economic environment (markets, service and infrastructure)
gives incentives or disincentives to select from these activities. Policy interventions lead to
changes in the socio-economic environment resulting in different (dis)incentives for the farm
households. The final outcome of the decision making process of the household is reflected in
the production pattern, productivity, social well being of the household and impact on
sustainability.
Therefore, the farm household framework can be used to assess the implications of different
policy measures for crop and technology choice, production, market exchange, labor use and
farm household welfare. Differences in risk behaviour (Roe and Graham-Tomasi, 1986),
market failures or missing markets (de Janry et al., 1991), and inter-temporal choice
(Fafchamps, 1993) can be also taken into account.
An adequate framework for the simultaneous appraisal of technological and economic options
for sustainable land use should take into account aspects of both supply and demand of
currently available and potentially new technologies. Therefore, a functional integration of
biophysical crop growth simulation models, programming models that reveal the resource
allocation implications of alternative crop and technologies choices and farm household
model that capture farmers` behavioural priorities, represents a major challenge.
4. Methodology
Economic models are used to identify the behavioural reasons for the choice of land use,
while agro-ecological models focus on the feasibility of technology and land use options for
specific agro-ecological conditions and on the assessment of their environmental
consequences. A combination of these two approaches could identify possible trade-offs
between economic and ecological objectives, assess the impact of policy interventions on
farm household decisions and the consequences for the economic performance of the
household and the natural resource conditions. Therefore, such a combined model has the
scope to assist policy making in an effective way (Ruben, Moll and Kuyvenhoven, 1998). The
development of an integrated approach to assist policy makers to promote a productive and
sustainable way of land management has to be based on 1) the choices between production
technologies and land use systems resulting from socio-economic factors (for example farm
household resources and objectives) and 2) an understanding of the biophysical processes (for
example crop growth in relation to input use, nutrient cycling).
Recently a new type of model, called bioeconomic models, have been developed which seem
to be appropriate to address the research objectives mentioned above. A bioeconomic model
combines socio-economic factors influencing farmers` objectives and constraints with
biophysical factors affecting production possibilities and the impacts of land management
practices. This approach is still in its infancy, but the initial results are promising (Barbier,
1996 and 1998. Kruseman, Hengsdijk, Ruben, Roebeling and Bade, 1997). Currently
available bioeconomic approaches consist of the following three components (Ruben, Moll
and Kuyvenhoven 1998):1
1. Mathematical programming models, that describe the actual behavior of individual
decision makers in the current situation and try to predict their behavior under
different policy scenarios;
2. Agro-ecological simulation models, that describe how different land use practices
affect yields and the condition of natural resources;
3. Aggregation procedures to address the effectiveness of policy interventions for
sustainable land use and well-being of the farmers at regional level.
In the following it is discussed in more detail how to design properly the three components of
the bioeconomic model for the predefined purpose.
4.1 Mathematical programming model
Econometric and programming models have been developed for the appraisal of land use and
for the analysis of the agricultural sector. The former model type is based on econometric
regressions or simultaneous equation systems explaining current land use pattern. Through
extrapolation of historical time series they can be used also for predictive purposes (Pindyck
and Rubinfeld 1991). The application of econometric models is criticized for the following
reasons (Berger 1999, Hazell and Norton 1985, Feder et al. 1985):
1
Ruben, Moll and Kuyvenhoven (1998) mentioned farm household models as the fourth component. Since this
model is the conceptual framework of the mathematical programming approach it is not listed as an extra
component here.
1. Data difficulties:

Large numbers of crops compete for the available fixed resources , and,
therefore, cross-supply effects are important components of the supply
function. Normally there are not enough degrees of freedom in a time series
data to estimate both own and cross-supply elasticities.

Aggregate time series on production are often quite unreliable in LDCs.

Inconsistent data base for the estimation of model coefficients and problems
with the statistical validation of parameters.

Measurement problems of variables like risk aversion or future expectations.
2. Rather simple economic models are considered to be characterized by price takers in
perfect competition with homogenous inputs and the “non-existence” of government
policies. Price effects may affect the progress and direction of the diffusion process of
innovations.
3. Differential adoption rates of technology by different economic groups and
institutional arrangements should be considered explicitly and in more detail.
4. Problems in capturing the consequences of changes in the economic structure. Policy
instruments need to take on values lying outside the range observed historically. This
possibility makes it unwise to base policy analyses on extrapolations from historically
estimated parameters.
A mathematical programming model helps to find the farm plan (defined by a set of activity
levels) that maximizes the objective function, but which does not violate any of the fixed
resource constraints, or involve any negative activity levels. They offer great possibilities to
formulate a wide range of actual and potential activities and to determine their relative
attractiveness. Advanced techniques offer the possibility to reflect farmers` behaviour
realistically, e.g. the inclusion of risk aversion and household food requirements in the
objective function and constraints (Hazell and Norton, 1985). Time considerations are quite
important in the context of decision-making processes in agriculture. Time span elapses
between the decision to carry out and the moment where the results of the process are
disposable. Farmers have to make decisions with different time horizons, the time span also
varies between the decision makers. There are some feedbacks to be considered between
short, medium and long term decisions. Feedbacks may be taken into account by using a
recursive structure.
Two extreme prototypes of agricultural sector models were defined by Hanf (1989). The
“simultaneous equilibrium approach” maximizes a common sectoral utility function and
assumes a perfect market mechanism. The “representative independent farm model” can be
defined as independently calculated representative farm models where the models are a
sample out of all farms. Computational results are added up to regional results. Hanf (1989)
concludes that the latter model should be chosen if the sector development is characterized by
1) imperfect markets, 2) behavior other than pure profit maximization and 3) adjustment
processes. It is taken into account that decisions with different time horizons have to be made
at the farm level. The approach offers the possibility to analyse the behaviour of the
individual farmer. Recursive and iterative procedures can be employed to guarantee certain
coordination within the sector development with respect to supply and demand of products
and production factors (Hanf and Pomarici, 1996).
Programming models are able to simulate adjustment of land use under changing conditions.
Therefore, they are an appropriate approach to analyze the choice among alternative activities
and technologies and to assess the impacts of alternative policies in the short and long run.
Brandes (1985) criticized linear programming methods in the sense that as a consequence of
compensating errors and due to the temptation of manipulation the model builder could give
the impression that his model reflects reality. An important weakness of these conventional
simulation models, apart from the aggregation error, is that they do not explicitly capture the
interactions between the farm households and therefore neglect transaction and information
costs.
Multi-agent Systems (MAS) can help to overcome these problems (Berger, 2000 and 2001).
This modeling method belongs to a new research area called Artificial Life. In social science
simulations, the agents usually represent humanlike individuals that interact within an
artificial society. MAS are computer systems set up by autonomously acting agents with
limited knowledge and information processing capacities maximizing their objective function
in an adaptive way. The autonomously acting agents in the model to be developed are
representative farm households. MAS consist of highly disaggregated farm programming
models with inter-household linkages. Mathematical programming models for each actor
represent the behavioural heterogeneity regarding production, investment, consumption and
marketing decisions. MAS can be based on the method of recursive linear programming,
similar to the representative independent farm model approach, to simulate individual
decision making over time. The complex decision making processes (e.g. expectation
formation, investment decision, production decision) of each actor can be decomposed in a
sequence of programming models.
Social interactions can be captured explicitly, including the communication concerning the
adoption of new technologies. The theory on diffusion of innovations emphasises the
importance of social networks for the adoption and diffusion of innovations (Rogers 1995,
Valente 1995). Rogers (1995) pointed out that typically the cumulative S-shaped adopter
distribution closely approach normality. The normal frequency distribution has several
characteristics that are useful in classifying adopters. Mean values and standard deviations are
used to classify adopters in the following four categories: early adopters, early majority, late
majority and laggards. Valente (1995 and 1996) introduced a method to estimate thresholds of
adoption based on the concept of social networks empirically. Empirically estimated
thresholds of individual farm households to adopt new technologies will help to illustrate the
diffusion patterns of promoted technologies. Figure 1 and Figure 2 confirm Rogers’
classification of adopter categories for a mosaic resistant cassava variety in the study area.
Figure 1: time of adoption compared to opinion leader
mosaic resistent cassava variety
20
Frequency
10
0
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
4,0
years
Figure 2: Cumulative number of adoptions
mosaic resistent cassava variety
60
50
40
30
20
10
0
-3
-2
-1
0
1
2
3
4
time of adoption compared to opinion leader
A farm household will compare his individual threshold with the present adoption level of his
social network and if the threshold is reached the farm’s net benefits from adoption will be
calculated. If the net benefit is positive the technology will be adopted. Standard economy
approaches predict that farmers might adjust “smoothly” to exogenous technical change.
Another scenario, which should be developed is the assumption that conservative farmers are
reluctant to technology adoption. If there is evidence that the behaviour of the farmers is
heterogeneous and interactions between the individual farmers play an important role for the
diffusion of innovations a model based on MAS seems to be the best solution.
4.2 Agro-ecological simulation model
The biophysical sub-model should determine the impact of various actual and potential land
management practices on crop yield and nutrient balances. Potential land management
practices are agricultural technologies promoted by CIAT in the study area. CIAT is
conducting different farm trials (inorganic fertilizer, organic fertilizer, soil and water
conservation measures) in close cooperation with at least 25 households. Soil and yield data
for the corresponding trials are available. The choice of an appropriate model, e.g. crop
growth simulation model (CAMASE, 1996. Williams, Jones and Dyke, 1984), technical
coefficient generator (Hengsdijk, Nieuwenhuyse and Bouman, 1998) or Artifical Neural
Networks (Bishop, 1995) is discussed in an interdisciplinary team at the moment.
4.3 Aggregation procedures and survey design
The aggregation procedures, the third component of the bioeconomic approach, with which
the effectiveness of policy intervention at regional level has to be addressed, is taken into
account within the survey design.
The following four steps for data collection were applied to meet the specific data
requirement of a bioeconomic multi-agent approach:
1. Household Survey (Round 1): Data collection for identification of representative
household types (n=106)
2. Identification of representative household types (Principal Component Analysis/Cluster
Analysis)
3. Household Survey (Round 2): Data collection for estimation of linear programming
parameters (n=20)
4. Generation of complete household data set (Monte-Carlo Simulation)
In the following these four steps are described more detailed:
4.3.1 Household Survey (Round 1):
The main objective of the first survey was to identify representative household types of two
villages in Mayuge district in eastern Uganda. Therefore, the procedure of stratified random
sampling was performed in order to reflect the proportion of non-trial households and trialhouseholds in the sampling universe. A listing indicated that 44 out of 608 households are
conducting agricultural technology trials in cooperation with CIAT and Africa 2000 Network
(A2N). Out of the first strata of 564 non-trial farmers 62 were selected randomly. Out of the
second strata of 44 trial farmers 5 were selected randomly for the subsequent analyses for
identification of representative household types. The other 39 trial farmers were also covered
in order to collect reliable data for technical coefficients of different technologies.
Descriptive statistics provide information on household characteristics (see appendix A1-A3),
household assets (see appendix A4), major crop types (see Figure 3), major information
sources for applied technologies (see Figure 4) and reasons for non-adoption of technologies
(see Figure 5).
Figure 3: Average proportion of cultivated
land under major crop types
60
%
land
40 beans
20
maize
cassava s.pot. coffee
bananas
other
gnuts
0
It should be emphasised here that peers and friends are after CIAT the most important
information sources for applied technologies (the importance of CIAT is not suprising since
they are very active in the study area). Furthermore, insufficient awareness, unavailability and
high input costs are the most important reasons for the non-adoption of new technologies.
These figures are an indicator for the importance of social networks for the adoption and
diffusion of innovations.
Figure 4: Major information sources for
applied technologies
CIAT/A2N
60
Peers/Fr.
40
Frequency
SG Relatives Other
Extension
DFI
20
0
Figure 5: Reasons for non-adoption of
technologies
insuf.aware
150
100
new
Frq.
not
other
available
input costs
50
0
4.3.2 Identification of representative household types:
A Principal Component Analysis was conducted for the following reasons: 1) to analyse the
structure of correlation among variables by defining a common set of underlying factors, 2) to
differentiate relevant from irrelevant variables for the subsequent Cluster Analysis (variables
which are not distinctive across the households can be eliminated). 3) the subsequent Cluster
Analysis can be conducted with uncorrelated factor scores (Backhaus, 1994). A Factor
Analysis reduces highly correlated variables to one factor. If highly correlated variables are
used for the subsequent Cluster Analysis, some characteristics have a stronger impact than
others when it comes to the clustering of objects. 4) Data can be reduced for the subsequent
Cluster Analysis (Hair,1998).
The disadvantage of this multivariate data analysis is certainly the associated loss of
information.
Various variables, captured in the first survey, were used as inputs for numerous Principal
Component Analyses. Because of their correlation structure and their relevance the following
8 variables were selected for the final Principal Component Analysis:
time of adoption compared to opinion leader, number of inorganic fertilizers/agrochemicals,
number of trial types conducted, number of different types of training, values of residence
buildings and other structures of the household, values of radios, walking time to output
market and percentage of quantity disposed on total production.
Different measures (e.g. Kaiser-Meyer-Olkin Measure of sampling adequacy, Bartlett Test of
Sphericity) indicated that this set of variables is appropriate for a Principal Component
Analysis. The most common criteria for the number of extracted factors is the eigenvalue.
Three factors have an eigenvalue greater than 1, explaining together 66,8 % of the variance
(see A5). Table 1 illustrates the rotated component matrix with the factor loadings. The
factors could be titled as: innovativeness, household assets and market orientation.
Table 1: Principal Component Analysis
Rotated Component Matrix a
1
number of inorganic
fertilizer/agrochemical
number of trial types
conducted
number of different
types of training
time of adoption
compared to opinion
leader
values of residence
buildings and other
structures of the
household
value of radios
walking time output
market(minutes)
% of quantity sold on
total production
Component
2
3
,90
,19
,19
,88
,06
,23
,84
,20
-,10
-,51
,25
,26
-,04
,84
-,04
,35
,77
-,10
,00
,07
,73
-,08
,22
-,68
Extrac tion Method: Principal Component Analys is.
Rotation Method: Varimax w ith Kais er Normalization.
a. Rotation converged in 6 iterations.
Computed factor scores were used as the input for the subsequent Cluster Analysis. The main
reasons for performing this type of multivariate analysis are 1) identification of homogenous
household groups, 2) to provide a criteria for the selection of households for the in-depth
interviews in the second household survey and 3) to provide a segmentation of the data base
for the generation of a complete household data set (for more details see 4.3.3).
Regarding the clustering algorithm a combination of hierachical and non-hierachical methods
was chosen in order to fine-tune the results and to have a validity check. No standard
procedure for the number of clusters to be formed exists. A simple example for a stopping
rule is to look on large increases in the agglomeration coefficient. A large increase indicates
that two very different clusters are being merged.
Finally, the following four clusters were identified (see Figure 5): subsistence farm
households (n=20), semi-subsistence farm households (n=35), commercial farm households
(n=7) and innovative trial farm households (n=5).
Figure 5: Farm-Household Groups
7
20
5
35
Semi-Subsistence Farm-Households
Subsistence Farm-Households
Commercial Farm-Households
Innovative Trial Farm-Households
Descriptive statistics illustrate the main differences between the different farm household
groups (see Table 2). The innovative trial farm households belong to the early adopters of a
mosaic resistant cassava variety, they are (by definition) the only group which is conducting
trials in cooperation with CIAT, they apply the highest number of inorganic fertilizers and
other agrochemicals and they are the only group with a reasonable number of different types
of agricultural training.
The commercial farm households have the highest mean values for the following variables:
value of residence and other structures of the household, value of agricultural equipment per
person involved in farming, total value of agricultural production, value of agricultural
production per acre cultivated land and quantity sold on total agricultural production.
Interesting is the fact that they belong to the late adopters of the mosaic resistant cassava
variety. There are two possible explanation: 1) cassava is not important as a cash crop and
therefore not of major interest of commercial farm households 2) wealthier households are
excluded from the communication process of the average farm households.
Subsistence and semi-subsistence farm households have relatively low mean values for the
following variables: years of schooling of household head, value of household assets, quantity
sold on total agricultural production, value of agricultural production (total and per acre
cultivated land) and number of inorganic fertilizers and other agrochemicals applied.
Furthermore, the subsistence farm households face very long walking times to the next output
market and belong to the group of late adopters.
Table 2: Characteristics of the identified clusters
Semi-Subsistence
Farm
Households
Household Characteristics
Years of schooling head
Years of schooling wife
Number of different types
of training (since 1990)
Household Assets
Value of residence and
other structures (Ush)
Value of radios (Ush)
Value of agricultural
equipment per person
involved in farming (Ush)
Crop Production
Total value of agricultural
production (Ush)
Value of agricultural
production per acre
cultivated land (Ush)
Quantity sold on total
production (%)
Walking time to output
market (minutes)
Intensity of land use2
Labor-land ratio3
Innovativeness
Time of adoption compared
to opinion leader (years)
Time of adoption compared
to personal network (%)
Number of technologies
adopted last 10 years
Number of trial types
conducted
Subsistence
Farm
Households
Commercial
Farm
Households
Innovative
Trial Farm
Households
4,4
4,3
0,7
5,5
3,3
0,3
12,4
8,1
1,0
7,7
5,6
4,2
837271
1267450
7601429
1951326
22143
5261
16500
4358
74286
9739
43682
5778
833102
455977
1634769
1065898
181956
182103
224206
206860
52
23
64
35
45
142
64
81
1,2
131
0,9
260
1,4
165
1,1
590
0,7
4,6
5,8
-2
0,41
0,66
0,73
0,33
5
5
6
8
0
0
0
7
4.3.3 Household Survey (Round 2):
The main objectives of the second survey are to collect data for the estimation of input-output
coefficients and farm income analysis. These data provide useful input for the calibration of
the programming matrices. Additionally, soil samples were taken and plots were measured to
provide more specific data for the subsequent modelling work.
2
intensity of land use is defined as ratio between land area cultivated last 12 months and total land size
3
labor-land ratio is defined as ratio between labor use on farm (person days) and cultivated land size
4.3.4 Generation of a complete household data set
For relevant policy analysis conclusions should be drawn on a regional level and not only for
single households. Therefore, the extrapolation of the results is a critical issue.
The households out of each cluster (preferably most closely to the cluster center) interviewed
in the second household survey are considered as representative for the corresponding cluster
regarding the input-output coefficients. Regarding the resource endowments (right-hand site
of the programming matrix) data of all the members of one cluster will be used calculate
mean values and standard deviations. Since these values and the size of each group are known
a Monte-Carlo Simulation is a useful tool to generate a complete household data set for the
sampling universe.
5. Conclusions for Bio-economic Modelling and Outlook
The preliminary results need further clarification but they seem to support the hypotheses that
social and personal networks, market orientation and factor endowment are central aspects to
understand and model the behavior and livelihood strategies of farm households in Eastern
Uganda. The bio-economic multi-agent approach will be applied in this study, because 1.)
interactions between individuals can be captured explicitly and 2.) simulations can predict the
short to long run impact of promoted technologies and policy interventions on household
welfare and condition of natural resources.
Completing the research studies will involve the following steps:
1. Calibration of the programming matrix for each cluster
2. Designing the “simulation matrix” with introduction of innovations and investment
planning for each cluster (Programming language C++)
3. Incorporation of biophysical sub-model in modelling framework
4. Run relevant policy scenarios (see research objectives)
Appendix
A1:
Descriptive Statistics: Household Characteristics
Min
number of household
members
number of
hh-members involved
in farming
age of household head
age of wife
years of schooling
household head
years of schooling wife
number of
hh-members
participated in training
(since1990)
number of
hh-members
participated in
extension in 1999/2000
Max
17
7,85
3,61
1
11
3,60
2,16
20
18
80
70
45,97
35,14
17,52
12,39
0
18
6,07
4,67
0
15
4,68
4,07
0
2
,36
,64
0
2
,22
,52
Primary Activity Household Head
other
9%
trading
10%
farming
72%
A3:
Primary Activity Wife
other
6%
farming
94%
Std. Dev.
1
A2:
wage
worker
9%
Mean
A4:
Descriptive Statistics: Household Asset Land
Min
total land size (owned
or operated, in acres)
land use intensitya
% of upland soils on
total land size
% of land inherited
% of land received as
a gift
% of land purchased
% of land under
freehold status
% of land under
customary status
Max
Mean
,02
30,00
5,61
5,26
,20
2,25
1,08
,60
0
100
79
23
0
100
34
45
0
100
11
31
0
100
39
45
0
100
72
43
0
100
20
39
a. rat io berween land area cultivated in 12 months and t otal land size
A5:
Total Variance Explained
Component
1
2
3
4
5
6
7
8
Std. Dev.
Initial Eigenvalues
% of
Cumulative
Total Variance
%
2,84
35,53
35,53
1,44
17,99
53,52
1,06
13,25
66,77
,91
11,33
78,10
,88
10,96
89,06
,46
5,78
94,84
,33
4,19
99,03
,08
,97
100,00
Extract ion M eth od: Principal Component A naly sis.
References
Agricultural Policy Secretariat Uganda (2000): Report on Characterization of Policies and
Institutions Affecting Land Management.
Backhaus, K., B.Erichson, W.Plinke and R.Weiber (1994): Multivariate Analysemethoden,
7.Auflage, Springer-Verlag, Berlin.
Baffoe, John K. (2000): Structural Adjustment and Agriculture in Uganda. International
Labour Organization, Working Paper No.149. Geneva.
Barbier, B. (1996): Impact of Market and Population Pressure on Production, Incomes and
Natural Resources in the Dryland Savannas of West Africa: Bioeconomic Modeling at
the Village Level. EPTD Discussion Paper No. 21. IFPRI, Washington.
Barbier, B. (1997): The Economic Determinants of Land Degradation in Developing
Countries.
Barbier, B. and G. Bergeron (1998): Natural Resource Management in the Hillsides Of
Honduras: Bioeconomic Modeling at the Micro-Watershed Level. EPTD Discussion
Paper No. 32. IFPRI, Washington.
Berger, T. (2000): Agentenbasierte räumliche Simulationsmodelle in der Landwirtschaft.
Anwendungsmöglichkeiten zur Bewertung von Diffusionsprozessen,
Ressourcennutzung und Politikoptionen. Agrarwirtschaft Sonderheft 168.
Berger, T. (2001): Agent-based Spatial Models Applied to Agriculture: A Simulation Tool for
Technology Diffusion, Resource Use Changes and Policy Analysis. Agricultural
Economics Volume 25, Issue 2, 01-September-2001.
Bishop, Christopher M. (1995): Neural Networks for Pattern Recognition, New York.
Brandes, W. (1985): Über die Grenzen der Schreibtischökonomie. Tübingen.
CAMASE (1996): Register of Agro-ecosystems Models. Ed.: Plentinger, M.C. and F.W.T.
Penning de Vries. Wageningen.
de Janvry, A., M. Fafchamps and E. Sadoulet (1991): Peasant Household Behaviour with
Missing Markets: Some Paradoxes Explained. In: The Economic Journal 101
Fafchamps, M. (1993): Sequential Labor Decisions under Uncertainty: An Estimable
Household Model of West African Farmers. Econometrica 61.
FAO (1999): Uganda Soil Fertitily Initiative, Draft Concept Paper. Rome.
Feder, G., R.E. Just and D. Zilberman (1985): Adoption of Agricultural Innovation in
Developing Countries: A Survey. In: Journal of Economic Development and Cultural
Change 33 (2).
Government of the Republic of Uganda (2000): Plan for Modernization pf Agriculture:
Eradicating Poverty in Uganda. Government Strategy and Operational Framework.
Hair, J.S., R.E. Anderson, R.L. Tatham, W.C. Black (1998): Multivariate Data Analysis, 5th
Edition, London.
Hanf, C.H. (1989): Agricultural Sector Analysis by Linear Prgramming Models. Kiel.
Hazell, P.B.R. and R.D. Norton (1985): Mathematical Prgramming for Economic Analysis in
Agriculture. New York.
Kruseman, G., H. Hengsdijk, R. Ruben, P. Roebeling, J. Bade (1997): Farm Household
Modelling System for the Analysis of Sustainable Land Use and Food Security:
Theoretical and Mathematical Description. DLV-Report No.7.
Land Degradation and Development (2000), Volume 11, Issue No. 2.
Pindyck, R.S. and D.L. Rubinfeld (1991): Econometric Models and Economic Forecasts. New
York.
Pomarici, E. and C.H. Hanf (1996): Outline of a Dynamic Mathematical Programming Model
for Regional Analysis of Agricultural Development. Materialsammlung zur
Agrarunternehmenslehre. Kiel.
Roe, T. and T. Graham-Tomasi (1986): Yield Risk in a Dynamic Model. In: Agricultural
Household Models: Extensions, Applications, and Policy, ed.: I. Singh, L. Squire and
J. Strauss. Baltimore and London.
Rogers, E. (1995): Diffusion of Innovations. 4.edition. New York.
Ruben, R., H. Menk and A. Kuyvenhoven (1998): Integrating Agricultural Research and
Policy Analysis: Analytical Framework and Policy Applications for Bio-economic
Modeling. In: Agricultural Systems, Vol. 58, No.3.
Sanchez, P.A., A.N. Izac, I. Valencia, and C. Pieri (1996): Soil Fertility Replenishment in
Africa. In: S.A. Breth (ed.): Achieving greater Impact from Research Investments in
Africa. Mexico City.
Singh, I., L. Squire and J. Strauss (1986): Agricultural Household Models: Extensions,
Applications, and Policy. Baltimore and London.
Stoorvogel, J.J. and E.M.A. Smaling (1990): Assessment of Soil nutrient Depletion in SubSaharan Africa: 1983-2000. Wageningen.
Valente, Thomas W. (1995): Network Models of the Diffusion of Innovations. Cresskill, New
Jersey.
Valente, T.W. (1996): Social Network Thresholds in the Diffusion of Innovations. In: Social
Networks 18 (1996) 69-89.
Vosti, S.A. and T. Reardon (1997): The Critical Triangle of Links among Sustainability,
Growth, and Poverty Alleviation. In: Vosti, S.A. and T. Reardon (ed.): Sustainability,
Growth and Poverty Alleviation. Baltimore and London.
Williams, J.R., C.A. Jones and P.T. Dyke (1984): A Modeling Approach to Determining the
Relationship between Erosion and Soil Productivity. Trans.ASAE 27.
World Bank (1993): Uganda Agriculture. Washington.
Wortmann, C.S. and C.K. Kaizzi (1998): Nutrient Balances and Expected Effects of
alternative Practices in Farming Systems of Uganda. In: Agric. Ecosystem.
Environ.115.129.